{-# LANGUAGE TemplateHaskell #-} {-# LANGUAGE CPP #-} {-# LANGUAGE OverloadedStrings #-} -- | This module uses template Haskell. Following is a simplified explanation of usage for those unfamiliar with calling Template Haskell functions. -- -- The function @embedFile@ in this modules embeds a file into the exceutable -- that you can use it at runtime. A file is represented as a @ByteString@. -- However, as you can see below, the type signature indicates a value of type -- @Q Exp@ will be returned. In order to convert this into a @ByteString@, you -- must use Template Haskell syntax, e.g.: -- -- > $(embedFile "myfile.txt") -- -- This expression will have type @ByteString@. Be certain to enable the -- TemplateHaskell language extension, usually by adding the following to the -- top of your module: -- -- > {-# LANGUAGE TemplateHaskell #-} module Data.FileEmbed ( -- * Embed at compile time embedFile , embedOneFileOf , embedDir , getDir -- * Embed as a IsString , embedStringFile , embedOneStringFileOf -- * Inject into an executable -- $inject #if MIN_VERSION_template_haskell(2,5,0) , dummySpace , dummySpaceWith #endif , inject , injectFile , injectWith , injectFileWith -- * Internal , stringToBs , bsToExp , strToExp ) where import Language.Haskell.TH.Syntax ( Exp (AppE, ListE, LitE, TupE, SigE, VarE) #if MIN_VERSION_template_haskell(2,5,0) , Lit (StringL, StringPrimL, IntegerL) #else , Lit (StringL, IntegerL) #endif , Q , runIO #if MIN_VERSION_template_haskell(2,7,0) , Quasi(qAddDependentFile) #endif ) import System.Directory (doesDirectoryExist, doesFileExist, getDirectoryContents) import Control.Exception (throw, ErrorCall(..)) import Control.Monad (filterM) import qualified Data.ByteString as B import qualified Data.ByteString.Char8 as B8 import Control.Arrow ((&&&), second) import Control.Applicative ((<$>)) import Data.ByteString.Unsafe (unsafePackAddressLen) import System.IO.Unsafe (unsafePerformIO) import System.FilePath (()) import Data.String (fromString) import Prelude as P -- | Embed a single file in your source code. -- -- > import qualified Data.ByteString -- > -- > myFile :: Data.ByteString.ByteString -- > myFile = $(embedFile "dirName/fileName") embedFile :: FilePath -> Q Exp embedFile fp = #if MIN_VERSION_template_haskell(2,7,0) qAddDependentFile fp >> #endif (runIO $ B.readFile fp) >>= bsToExp -- | Embed a single existing file in your source code -- out of list a list of paths supplied. -- -- > import qualified Data.ByteString -- > -- > myFile :: Data.ByteString.ByteString -- > myFile = $(embedFile' [ "dirName/fileName", "src/dirName/fileName" ]) embedOneFileOf :: [FilePath] -> Q Exp embedOneFileOf ps = (runIO $ readExistingFile ps) >>= \ ( path, content ) -> do #if MIN_VERSION_template_haskell(2,7,0) qAddDependentFile path #endif bsToExp content where readExistingFile :: [FilePath] -> IO ( FilePath, B.ByteString ) readExistingFile xs = do ys <- filterM doesFileExist xs case ys of (p:_) -> B.readFile p >>= \ c -> return ( p, c ) _ -> throw $ ErrorCall "Cannot find file to embed as resource" -- | Embed a directory recursively in your source code. -- -- > import qualified Data.ByteString -- > -- > myDir :: [(FilePath, Data.ByteString.ByteString)] -- > myDir = $(embedDir "dirName") embedDir :: FilePath -> Q Exp embedDir fp = do typ <- [t| [(FilePath, B.ByteString)] |] e <- ListE <$> ((runIO $ fileList fp) >>= mapM (pairToExp fp)) return $ SigE e typ -- | Get a directory tree in the IO monad. -- -- This is the workhorse of 'embedDir' getDir :: FilePath -> IO [(FilePath, B.ByteString)] getDir = fileList pairToExp :: FilePath -> (FilePath, B.ByteString) -> Q Exp pairToExp _root (path, bs) = do #if MIN_VERSION_template_haskell(2,7,0) qAddDependentFile $ _root ++ '/' : path #endif exp' <- bsToExp bs return $! TupE [LitE $ StringL path, exp'] bsToExp :: B.ByteString -> Q Exp #if MIN_VERSION_template_haskell(2, 5, 0) bsToExp bs = return $ VarE 'unsafePerformIO `AppE` (VarE 'unsafePackAddressLen `AppE` LitE (IntegerL $ fromIntegral $ B8.length bs) #if MIN_VERSION_template_haskell(2, 8, 0) `AppE` LitE (StringPrimL $ B.unpack bs)) #else `AppE` LitE (StringPrimL $ B8.unpack bs)) #endif #else bsToExp bs = do helper <- [| stringToBs |] let chars = B8.unpack bs return $! AppE helper $! LitE $! StringL chars #endif stringToBs :: String -> B.ByteString stringToBs = B8.pack -- | Embed a single file in your source code. -- -- > import Data.String -- > -- > myFile :: IsString a => a -- > myFile = $(embedStringFile "dirName/fileName") -- -- Since 0.0.9 embedStringFile :: FilePath -> Q Exp embedStringFile fp = #if MIN_VERSION_template_haskell(2,7,0) qAddDependentFile fp >> #endif (runIO $ P.readFile fp) >>= strToExp -- | Embed a single existing string file in your source code -- out of list a list of paths supplied. -- -- Since 0.0.9 embedOneStringFileOf :: [FilePath] -> Q Exp embedOneStringFileOf ps = (runIO $ readExistingFile ps) >>= \ ( path, content ) -> do #if MIN_VERSION_template_haskell(2,7,0) qAddDependentFile path #endif strToExp content where readExistingFile :: [FilePath] -> IO ( FilePath, String ) readExistingFile xs = do ys <- filterM doesFileExist xs case ys of (p:_) -> P.readFile p >>= \ c -> return ( p, c ) _ -> throw $ ErrorCall "Cannot find file to embed as resource" strToExp :: String -> Q Exp #if MIN_VERSION_template_haskell(2, 5, 0) strToExp s = return $ VarE 'fromString `AppE` LitE (StringL s) #else strToExp s = do helper <- [| fromString |] return $! AppE helper $! LitE $! StringL s #endif notHidden :: FilePath -> Bool notHidden ('.':_) = False notHidden _ = True fileList :: FilePath -> IO [(FilePath, B.ByteString)] fileList top = fileList' top "" fileList' :: FilePath -> FilePath -> IO [(FilePath, B.ByteString)] fileList' realTop top = do allContents <- filter notHidden <$> getDirectoryContents (realTop top) let all' = map ((top ) &&& (\x -> realTop top x)) allContents files <- filterM (doesFileExist . snd) all' >>= mapM (liftPair2 . second B.readFile) dirs <- filterM (doesDirectoryExist . snd) all' >>= mapM (fileList' realTop . fst) return $ concat $ files : dirs liftPair2 :: Monad m => (a, m b) -> m (a, b) liftPair2 (a, b) = b >>= \b' -> return (a, b') magic :: B.ByteString -> B.ByteString magic x = B8.concat ["fe", x] sizeLen :: Int sizeLen = 20 getInner :: B.ByteString -> B.ByteString getInner b = let (sizeBS, rest) = B.splitAt sizeLen b in case reads $ B8.unpack sizeBS of (i, _):_ -> B.take i rest [] -> error "Data.FileEmbed (getInner): Your dummy space has been corrupted." padSize :: Int -> String padSize i = let s = show i in replicate (sizeLen - length s) '0' ++ s #if MIN_VERSION_template_haskell(2,5,0) -- | Allocate the given number of bytes in the generate executable. That space -- can be filled up with the 'inject' and 'injectFile' functions. dummySpace :: Int -> Q Exp dummySpace = dummySpaceWith "MS" -- | Like 'dummySpace', but takes a postfix for the magic string. In -- order for this to work, the same postfix must be used by 'inject' / -- 'injectFile'. This allows an executable to have multiple -- 'ByteString's injected into it, without encountering collisions. -- -- Since 0.0.8 dummySpaceWith :: B.ByteString -> Int -> Q Exp dummySpaceWith postfix space = do let size = padSize space magic' = magic postfix start = B8.unpack magic' ++ size magicLen = B8.length magic' len = magicLen + sizeLen + space chars = LitE $ StringPrimL $ #if MIN_VERSION_template_haskell(2,6,0) map (toEnum . fromEnum) $ #endif start ++ replicate space '0' [| getInner (B.drop magicLen (unsafePerformIO (unsafePackAddressLen len $(return chars)))) |] #endif -- | Inject some raw data inside a @ByteString@ containing empty, dummy space -- (allocated with @dummySpace@). Typically, the original @ByteString@ is an -- executable read from the filesystem. inject :: B.ByteString -- ^ bs to inject -> B.ByteString -- ^ original BS containing dummy -> Maybe B.ByteString -- ^ new BS, or Nothing if there is insufficient dummy space inject = injectWith "MS" -- | Like 'inject', but takes a postfix for the magic string. -- -- Since 0.0.8 injectWith :: B.ByteString -- ^ postfix of magic string -> B.ByteString -- ^ bs to inject -> B.ByteString -- ^ original BS containing dummy -> Maybe B.ByteString -- ^ new BS, or Nothing if there is insufficient dummy space injectWith postfix toInj orig = if toInjL > size then Nothing else Just $ B.concat [before, magic', B8.pack $ padSize toInjL, toInj, B8.pack $ replicate (size - toInjL) '0', after] where magic' = magic postfix toInjL = B.length toInj (before, rest) = B.breakSubstring magic' orig (sizeBS, rest') = B.splitAt sizeLen $ B.drop (B8.length magic') rest size = case reads $ B8.unpack sizeBS of (i, _):_ -> i [] -> error $ "Data.FileEmbed (inject): Your dummy space has been corrupted. Size is: " ++ show sizeBS after = B.drop size rest' -- | Same as 'inject', but instead of performing the injecting in memory, read -- the contents from the filesystem and write back to a different file on the -- filesystem. injectFile :: B.ByteString -- ^ bs to inject -> FilePath -- ^ template file -> FilePath -- ^ output file -> IO () injectFile = injectFileWith "MS" -- | Like 'injectFile', but takes a postfix for the magic string. -- -- Since 0.0.8 injectFileWith :: B.ByteString -- ^ postfix of magic string -> B.ByteString -- ^ bs to inject -> FilePath -- ^ template file -> FilePath -- ^ output file -> IO () injectFileWith postfix inj srcFP dstFP = do src <- B.readFile srcFP case injectWith postfix inj src of Nothing -> error "Insufficient dummy space" Just dst -> B.writeFile dstFP dst {- $inject The inject system allows arbitrary content to be embedded inside a Haskell executable, post compilation. Typically, file-embed allows you to read some contents from the file system at compile time and embed them inside your executable. Consider a case, instead, where you would want to embed these contents after compilation. Two real-world examples are: * You would like to embed a hash of the executable itself, for sanity checking in a network protocol. (Obviously the hash will change after you embed the hash.) * You want to create a self-contained web server that has a set of content, but will need to update the content on machines that do not have access to GHC. The typical workflow use: * Use 'dummySpace' or 'dummySpaceWith' to create some empty space in your executable * Use 'injectFile' or 'injectFileWith' from a separate utility to modify that executable to have the updated content. The reason for the @With@-variant of the functions is for cases where you wish to inject multiple different kinds of content, and therefore need control over the magic key. If you know for certain that there will only be one dummy space available, you can use the non-@With@ variants. -}